An intersection cooperative control system and method based on visual semantic second-reading analysis and dynamic risk locking

Abstract

The application relates to a kind of intersection cooperative control system and method.The system collects red light image in real time through vehicle-mounted camera, uses OCR to extract countdown value, and introduces interframe compensation algorithm to carry out timing prediction on damaged identification result, to ensure the continuity of perception data.The system calculates theoretical traffic strategy based on countdown data, and simultaneously starts dynamic risk locking mode, identifies the latent risk in blind area by scanning the "overflow risk area" of the edge of the intersection obstruction and the moving shadow vector, and executes safety arbitration when detecting the collision trend of blind area sudden crossing, to realize the double arbitration of safety and efficiency by sending high-priority message through CAN / CAN-FD bus to cut off power output.The application does not depend on high-precision map and network API, and significantly improves the safety prediction ability and system robustness of intersection traffic.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving assistance system technology, specifically to a comprehensive decision-making system that combines traffic light countdown, vehicle dynamics calculation, and dynamic obstacle detection in blind spots at intersections, and is particularly suitable for safe passage control at intersections in L2+ and L3 level autonomous driving scenarios. Background Technology

[0002] Among existing intelligent driving assistance technologies, green wave speed guidance systems have become increasingly common. These systems typically calculate the recommended speed for vehicles passing through intersections based on navigation data and traffic light information, prompting drivers to accelerate or decelerate to improve traffic flow. However, existing green wave guidance technologies present significant safety hazards: The blind pursuit of efficiency: Existing systems (such as in-vehicle navigation and GLOSA systems) primarily focus on matching "time windows." When a green light is about to end (e.g., 3 seconds remaining), the system often calculates a "speed through" suggestion based on physical kinematics formulas. This calculation only considers whether the vehicle can cross the stop line before the red light turns on, completely ignoring the dynamic environment inside the intersection. Risk of sudden crossing from blind spots: At complex urban intersections, pedestrians or non-motorized vehicles often wait in blind spots (such as in front of buses or behind green belts) in the cross lanes. When the vehicle accelerates to run a red light as advised by the system, if a pedestrian suddenly runs a red light (i.e., suddenly crossing from a blind spot), the vehicle is accelerating and the driver's attention is focused on the traffic light, which can easily lead to a serious traffic accident due to insufficient reaction time or a longer braking distance.

[0003] Technological disconnect: Current Automatic Emergency Braking (AEB) and green wave guidance systems typically operate independently. The green wave system is responsible for "acceleration," while AEB is responsible for "emergency braking." The lack of logical coordination between the two leads to a disconnect in the driving experience and can even cause control conflicts in extreme situations.

[0004] Therefore, there is an urgent need for a method for making traffic decisions at intersections that can conduct a "safety pre-check" before suggesting acceleration and have a "dynamic veto power" during the acceleration process. Summary of the Invention

[0005] The purpose of this invention is to provide a method for intersection traffic decision-making and interaction based on dynamic risk locking, which aims to solve the safety hazards caused by the neglect of sudden risks at intersections in existing green wave guidance systems. The core logic of this invention lies in establishing a "dual arbitration mechanism for efficiency and safety." The system not only calculates "whether it can pass" (time feasibility), but also "whether it dares to pass" (environmental safety). The specific technical solution includes the following steps: The system identifies the countdown number (T_light) of the forward traffic light through an on-board camera, and calculates the time required for the vehicle to reach the intersection conflict area (t_arrive) and the time required for the vehicle to completely pass through the intersection (t_pass) by combining the current distance (S) between the vehicle and the intersection and the vehicle speed (V). If t_pass < T_light - safety threshold, the system initially generates a "suggest acceleration" passing strategy. "The present invention introduces an 'inter-frame compensation algorithm' into traffic light recognition. Due to road surface jitter or occlusion, OCR may not be able to read the number accurately at a certain moment. Instead of reporting an error directly, the system activates a prediction module: for example, if the T-1 frame is recognized as '5' and the T-2 frame is recognized as '6', the system can logically infer that the current number is '4' even if the T frame is damaged. This independent perception capability that is separated from high-definition maps and network APIs ensures the viability of the system in tunnel entrances, remote areas and environments with network fluctuations." While generating the "suggest acceleration" strategy, the system activates the "dynamic risk locking mode". This mode uses an on-board wide-angle camera and side radar to perform high-frequency scanning specifically for the "potential conflict areas" on both sides of the intersection. The system identifies obstructions (such as buses, guardrails) through a semantic segmentation algorithm, and locks the "overflow risk area" at the edge of the obstruction. If limb features or movement trends of dynamic objects (pedestrians, non-motor vehicles) are detected in the risk area, the area is determined to be in a "high-risk state". The system performs safety arbitration on the "suggest acceleration" strategy: Scenario A (clear): if there is no dynamic object in the risk area, the "suggest acceleration" instruction is maintained, and a green passing arrow is displayed on the instrument panel. Scenario B (potential risk): if a dynamic object is detected in the risk area, regardless of whether the remaining time of the traffic light is sufficient, the system immediately rejects the acceleration suggestion and generates a "defensive deceleration" instruction. A red warning shield is displayed on the instrument panel, with the prompt "There is a person in the blind area, do not rush through". If the driver has responded to the acceleration suggestion (increased throttle opening), but the system suddenly detects that an object rushes out of the risk area quickly (such as a pedestrian running suddenly), the system will trigger "dynamic veto": Level 1 intervention: immediately cut off power output, pre-charge the brake system, and emit a rapid alarm sound. Level 2 intervention: if the driver does not release the accelerator or reacts slowly, the system actively intervenes in braking (AEB) to ensure that the vehicle stops before the conflict area. The following is a detailed description of the implementation process of the present invention in combination with a specific urban intersection scenario. Scenario setting: the vehicle is traveling on an urban main road, 120 meters away from the forward crossroads. The current vehicle speed is 50km / h (about 13.9m / s). Visual recognition: the on-board forward-looking camera captures that the forward signal light is green, and the countdown number shows "4" (that is, 4 seconds remaining). Kinematic calculation: Time required for the vehicle to reach the stop line: t_to = 120m / 13.9m / s ≈ 8.6 seconds. Clearly, maintaining the current speed will prevent them from passing through during the green light period. The system calculates the acceleration plan: if accelerating to 65km / h (approximately 18m / s), the arrival time becomes 6.6 seconds. Considering the intersection width and the yellow light duration (usually 3 seconds), the system determines that if accelerating immediately, it is theoretically possible to pass through the intersection before the red light completely ends. Preliminary decision: The system initially generates a strategy of "suggesting acceleration to 65km / h to pass through". Before issuing the acceleration prompt, the system did not execute it immediately, but instead started the "dynamic risk lock" program: blind spot scanning: the system retrieved data from the right-side wide-angle camera and identified a bus that was starting up on the right side of the intersection, blocking the view of the pedestrian crossing. Feature extraction: The system identified the legs and moving shadows of pedestrians within the "overflow risk zone" at the edge of the bus front (indicating that someone was preparing to cross quickly). Risk assessment: The system determines that this area is in a "high-risk state" with an extremely high probability of sudden blind spot crossing. Logical rejection: Although time allows passage, the security arbitration module determines that "acceleration risk > passage benefit". Command correction: The system cancels the "suggested acceleration" command and instead issues a "defensive deceleration" command. Human-computer interaction: Vision: The green arrow that was originally going to be displayed on the dashboard HUD has disappeared, and has been replaced by a flashing red shield icon, highlighting the right blind spot. Voice prompt: "Pedestrian detected in the right blind spot. Acceleration is prohibited. Please prepare to brake and slow down." If the driver ignores the voice prompt and insists on pressing the accelerator pedal to run a red light: Status monitoring: The system detects that the accelerator pedal opening exceeds 80% and the vehicle speed rapidly increases to 60km / h. Mutation detection: At this moment, a pedestrian suddenly rushed out from the right side of the bus (with a relatively fast lateral speed). Dynamic intervention: The system determines that the collision risk is imminent. T+0ms: The system forcibly cuts off the engine / motor torque output. T+100ms: The system automatically establishes hydraulic braking, and the vehicle begins to decelerate. Result: The vehicle came to a smooth stop 2 meters away from the pedestrian, averting a serious accident. Beneficial effects Compared with the prior art, the present invention has the following significant advantages: 1. Extremely high perception robustness and continuous compensation mechanism: This invention effectively solves the visual interference problem faced by vehicle-mounted cameras in intersection environments through an inter-frame compensation algorithm. Even when road bumps cause image blurring, or large vehicles momentarily obscure the traffic light countdown, the system can still use time-series prediction logic to "blindly fill in" missing data, ensuring the continuity of traffic strategy calculation and avoiding frequent system starts and stops or false alarms caused by signal interruptions. 2. Proactive safety defense and prediction capabilities for sudden crossings in blind spots: Unlike traditional obstacle recognition, this invention innovatively proposes "overflow risk zone" scanning and moving shadow vector analysis. The system can predict the tendency of traffic participants (such as pedestrians and non-motorized vehicles) to rush out before they fully appear in the lateral blind spot by observing changes in ground shadows. This "beyond visual range" warning capability significantly reduces the collision risk when turning right or going straight at intersections. 3. A Dynamic Arbitration Logic for Safety and Efficiency: The system not only provides green wave traffic guidance to improve intersection efficiency, but also establishes a "safety veto" mechanism. Through dual decision-making nodes, the system can determine the conflict between driver intent and objective risks in real time. When abnormal acceleration by a driver in a dangerous state is detected, physical intervention can be implemented through high-priority instructions, pushing the safety boundary to the highest level while ensuring traffic efficiency. 4. Low-cost and highly independent deployment advantages: This invention is entirely based on in-vehicle visual perception and local algorithms, without relying on high-precision maps, V2X vehicle-to-infrastructure roadside units, or expensive network API interfaces. This means that the system can maintain full functionality even at intersections with poor infrastructure, tunnel entrances, or in environments with weak networks, greatly reducing the cost threshold for large-scale mass production and promotion. 5. Precise Human-Machine Collaboration and Tiered Intervention Strategy: The system provides a progressive interaction logic from "HUD visual guidance" to "voice warning" and then to "forced power cut-off." This tiered intervention mode respects the driver's control while providing algorithmic safeguards in extremely dangerous situations, effectively mitigating the control conflict between the autonomous driving / assisted driving system and the human driver. 6. In the physical execution of the fourth stage described above, the system sends a control message with the highest priority to the power control unit via the vehicle's CAN or CAN-FD bus. This message overrides the driver's accelerator pedal sensor signal. For new energy vehicles, the system instructs the inverter to directly set the motor torque output request to zero; for gasoline vehicles, power is cut off through electronic throttle or fuel injection pulse control, thereby forcibly preventing vehicle acceleration in the event of driver error. Attached Figure Description Figure 1 This is a schematic diagram of the system hardware composition; Figure 2 This is the core algorithm logic flowchart; Figure 3 This is a diagram illustrating the principle of inter-frame compensation. Figure 4 This is a diagram illustrating the principle of blind spot risk identification. Explanation of reference numerals in the attached figures: 1. Vehicle-mounted camera; 2. Radar sensor; 3. V2X communication module; 4. Electronic control unit (ECU); 5. Power actuator; 6. Braking actuator; 201 inter-frame compensation module; 301. This vehicle, 302. Bus, 303. Blind spot vehicle area, a. Effective perception area, b. Dynamic risk spillover area. Detailed Implementation Figure 1 is a schematic diagram of the system hardware composition in an embodiment of the present invention. As shown in Figure 1, the system includes: an onboard camera 1, a radar sensor 2, a V2X communication module 3, an electronic control unit (ECU) 4, a power actuator 5, and a braking actuator 6. The system is divided into different response zones based on distance and risk level: Area A (farthest): corresponds to the "reminder" strategy; Zone B (Mid-range): Corresponds to the "Alarm" policy; Zone C (near distance / blind spot): corresponds to the "sudden stop" strategy. Figure 2 is a flowchart of the core algorithm logic in an embodiment of the present invention. As shown in Figure 2, the core algorithm logic includes the following steps: Step 101 [System Startup and Initialization]: The system completes self-test, activates the forward and side-view sensing units, and enters standby listening state. Step 102 [Real-time Acquisition of Multi-Source Sensing Data]: Real-time acquisition of traffic light images and side blind spot image streams at the intersection via vehicle-mounted camera 1. Step 103 [Visual Semantic Analysis and Inter-Frame Compensation]: Use OCR to extract the countdown value and simultaneously start the inter-frame compensation algorithm to perform temporal prediction compensation for the current damaged or blurred recognition results by combining historical frame data. Step 104 [Theoretical Passage Window and Strategy Calculation]: Based on the analytical countdown and vehicle dynamics model, calculate the theoretical recommended speed range for vehicles to safely pass through the intersection. Step 105 [Blind Spot and Dynamic Risk Scanning at Intersections]: The system performs high-frequency scanning of the "overflow risk zone" at the edge of obstructions and performs vector change analysis on moving shadows. Step 106 [Safety Arbitration Determination]: Diamond Decision Node. Determine whether there are dynamic obstacles with a collision tendency within the "overflow risk zone". If no (N), the environment is safe, proceed to step 108; if yes (Y), there is a risk, proceed to step 107. Step 107 [Security Lock and Command Correction]: The system executes a "one-vote veto" mechanism, immediately cancels or blocks acceleration suggestions, switches the interactive interface to a red warning shield and prompts defensive deceleration. Step 108 [Output Green Wave Traffic Guidance Instruction]: After confirming that the environment is clear, output a green acceleration passage suggestion to the driver or maintain the current guidance strategy. Step 109 [Driver Intent Conflict Monitoring]: Diamond Decision Node. Under risk conditions, monitor in real time whether the driver ignores the warning and performs abnormal acceleration behaviors such as pressing the accelerator pedal deeply. If yes (Y), a control conflict is detected, proceed to step 110; if no (N), the driver responds to the warning, proceed to step 111. Step 110 [Highest Priority Power Intervention Execution]: The system sends an overlay-level message via the CAN / CAN-FD bus to forcibly cut off power or reduce torque to zero, thus forcibly preventing the vehicle from accelerating through the traffic lights. Step 111 [Continuous Security Monitoring State]: The system maintains defensive alerts and automatically returns to step 102 after execution to start the real-time scanning loop for the next frame of data. Figure 3 is a schematic diagram of the inter-frame compensation principle in an embodiment of the present invention. As shown in Figure 3, the inter-frame compensation module 201 is communicatively connected to the image acquisition unit. When the visual recognition confidence at the current time T is lower than a preset threshold, the inter-frame compensation module 201 retrieves the historical recognition values ​​(7, 6, and 5) recorded in the memory at times T-3, T-2, and T-1, calculates the logical compensation value of 4 at the current time through linear subtraction logic, and outputs it to the subsequent passage strategy calculation unit. Figure 4 is a schematic diagram of the blind spot risk identification principle in an embodiment of the present invention. As shown in Figure 4, 301 is the vehicle itself, 302 is the bus, and 303 is the vehicle in the blind spot. Region a is the effective perception area of ​​this vehicle. Within this area, the image acquisition unit can directly acquire the visual features of traffic participants and perform routine target detection. Area b is a dynamic risk spillover zone, i.e., a visual blind spot formed by large obstructions (302). This invention detects the risk of running away by monitoring pixel changes (such as moving shadow vectors) at the edge of area b before traffic participants leave the blind spot and enter area a.

Claims

1. A method for intersection traffic decision-making and interaction based on dynamic risk locking, characterized in that, Includes the following steps: Step 1: Use an onboard camera to collect video streams of the intersection and extract the countdown character features of the traffic lights; the perception process has inter-frame compensation logic, that is, when a single frame recognition fails, the current frame is temporally predicted and compensated according to the numerical jump rules of the preceding consecutive frames to ensure the continuity of the perception data. Step 2: Based on the countdown data and vehicle dynamics model, calculate the theoretical recommended speed and traffic strategy for the vehicle passing through the intersection; Step 3: Before outputting the traffic strategy, scan the blind spots and potential conflict areas of the intersection using onboard sensors to detect the presence of dynamic obstacles; Step 4: If a dynamic obstacle is detected and there is a risk of collision, the passage strategy is rejected or modified, and a defensive driving instruction is generated.

2. The method according to claim 1, characterized in that, The "dynamic obstacles" in step 3 include pedestrians or non-motorized vehicles that are located behind obstructions to the view at the intersection and have a tendency to enter the lane.

3. The method according to claim 1, characterized in that, The "correct traffic strategy" in step 4 specifically includes: canceling the acceleration suggestion and generating voice and visual prompts for deceleration and braking.

4. The method according to claim 1, characterized in that, It also includes step 5: If a sudden risk is detected, the system sends a control message with the highest priority to the power control unit through the vehicle CAN or CAN-FD bus. This message can override the driver's accelerator pedal request and force the power cut-off or torque zeroing operation.

5. The method according to claim 1, characterized in that, The system employs a dynamic adjustment mechanism for "safety thresholds": when blind spots at intersections are severely obstructed, the risk detection range is automatically expanded to increase the sensitivity of rejection acceleration.

Images